- 31 Jan, 2017 33 commits
-
-
David Gibson authored
This updates the KVM_CAP_SPAPR_RESIZE_HPT capability to advertise the presence of in-kernel HPT resizing on KVM HV. Signed-off-by:
David Gibson <david@gibson.dropbear.id.au> Signed-off-by:
Paul Mackerras <paulus@ozlabs.org>
-
David Gibson authored
This adds the "guts" of the implementation for the HPT resizing PAPR extension. It has the code to allocate and clear a new HPT, rehash an existing HPT's entries into it, and accomplish the switchover for a KVM guest from the old HPT to the new one. Signed-off-by:
-
David Gibson authored
This adds a not yet working outline of the HPT resizing PAPR extension. Specifically it adds the necessary ioctl() functions, their basic steps, the work function which will handle preparation for the resize, and synchronization between these, the guest page fault path and guest HPT update path. The actual guts of the implementation isn't here yet, so for now the calls will always fail. Signed-off-by:
-
David Gibson authored
The kvm_unmap_rmapp() function, called from certain MMU notifiers, is used to force all guest mappings of a particular host page to be set ABSENT, and removed from the reverse mappings. For HPT resizing, we will have some cases where we want to set just a single guest HPTE ABSENT and remove its reverse mappings. To prepare with this, we split out the logic from kvm_unmap_rmapp() to evict a single HPTE, moving it to a new helper function. Signed-off-by:
-
David Gibson authored
The KVM_PPC_ALLOCATE_HTAB ioctl() is used to set the size of hashed page table (HPT) that userspace expects a guest VM to have, and is also used to clear that HPT when necessary (e.g. guest reboot). At present, once the ioctl() is called for the first time, the HPT size can never be changed thereafter - it will be cleared but always sized as from the first call. With upcoming HPT resize implementation, we're going to need to allow userspace to resize the HPT at reset (to change it back to the default size if the guest changed it). So, we need to allow this ioctl() to change the HPT size. This patch also updates Documentation/virtual/kvm/api.txt to reflect the new behaviour. In fact the documentation was already slightly incorrect since 572abd56 "KVM: PPC: Book3S HV: Don't fall back to smaller HPT size in allocation ioctl" Signed-off-by:
-
David Gibson authored
Currently, kvmppc_alloc_hpt() both allocates a new hashed page table (HPT) and sets it up as the active page table for a VM. For the upcoming HPT resize implementation we're going to want to allocate HPTs separately from activating them. So, split the allocation itself out into kvmppc_allocate_hpt() and perform the activation with a new kvmppc_set_hpt() function. Likewise we split kvmppc_free_hpt(), which just frees the HPT, from kvmppc_release_hpt() which unsets it as an active HPT, then frees it. We also move the logic to fall back to smaller HPT sizes if the first try fails into the single caller which used that behaviour, kvmppc_hv_setup_htab_rma(). This introduces a slight semantic change, in that previously if the initial attempt at CMA allocation failed, we would fall back to attempting smaller sizes with the page allocator. Now, we try first CMA, then the page allocator at each size. As far as I can tell this change should be harmless. To match, we make kvmppc_free_hpt() just free the actual HPT itself. The call to kvmppc_free_lpid() that was there, we move to the single caller. Signed-off-by:
-
David Gibson authored
Currently the kvm_hpt_info structure stores the hashed page table's order, and also the number of HPTEs it contains and a mask for its size. The last two can be easily derived from the order, so remove them and just calculate them as necessary with a couple of helper inlines. Signed-off-by:
Thomas Huth <thuth@redhat.com> Signed-off-by:
-
David Gibson authored
Currently, the powerpc kvm_arch structure contains a number of variables tracking the state of the guest's hashed page table (HPT) in KVM HV. This patch gathers them all together into a single kvm_hpt_info substructure. This makes life more convenient for the upcoming HPT resizing implementation. Signed-off-by:
-
David Gibson authored
The difference between kvm_alloc_hpt() and kvmppc_alloc_hpt() is not at all obvious from the name. In practice kvmppc_alloc_hpt() allocates an HPT by whatever means, and calls kvm_alloc_hpt() which will attempt to allocate it with CMA only. To make this less confusing, rename kvm_alloc_hpt() to kvm_alloc_hpt_cma(). Similarly, kvm_release_hpt() is renamed kvm_free_hpt_cma(). Signed-off-by:
-
David Gibson authored
This adds a new powerpc-specific KVM_CAP_SPAPR_RESIZE_HPT capability to advertise whether KVM is capable of handling the PAPR extensions for resizing the hashed page table during guest runtime. It also adds definitions for two new VM ioctl()s to implement this extension, and documentation of the same. Note that, HPT resizing is already possible with KVM PR without kernel modification, since the HPT is managed within userspace (qemu). The capability defined here will only be set where an in-kernel implementation of resizing is necessary, i.e. for KVM HV. To determine if the userspace resize implementation can be used, it's necessary to check KVM_CAP_PPC_ALLOC_HTAB. Unfortunately older kernels incorrectly set KVM_CAP_PPC_ALLOC_HTAB even with KVM PR. If userspace it want to support resizing with KVM PR on such kernels, it will need a workaround. Signed-off-by:
-
David Gibson authored
Both KVM_CREATE_SPAPR_TCE_64 and KVM_REINJECT_CONTROL have section number 4.98 in Documentation/virtual/kvm/api.txt, presumably due to a naive merge. This corrects the duplication. [paulus@ozlabs.org - correct section numbers for following sections, KVM_PPC_CONFIGURE_V3_MMU and KVM_PPC_GET_RMMU_INFO, as well.] Signed-off-by:
-
Paul Mackerras authored
This merges in the POWER9 radix MMU host and guest support, which was put into a topic branch because it touches both powerpc and KVM code. Signed-off-by: -
Paul Mackerras authored
This adds a few last pieces of the support for radix guests: * Implement the backends for the KVM_PPC_CONFIGURE_V3_MMU and KVM_PPC_GET_RMMU_INFO ioctls for radix guests * On POWER9, allow secondary threads to be on/off-lined while guests are running. * Set up LPCR and the partition table entry for radix guests. * Don't allocate the rmap array in the kvm_memory_slot structure on radix. * Don't try to initialize the HPT for radix guests, since they don't have an HPT. * Take out the code that prevents the HV KVM module from initializing on radix hosts. At this stage, we only support radix guests if the host is running in radix mode, and only support HPT guests if the host is running in HPT mode. Thus a guest cannot switch from one mode to the other, which enables some simplifications. Signed-off-by:
Michael Ellerman <mpe@ellerman.id.au>
-
Paul Mackerras authored
On POWER9 DD1, we need to invalidate the ERAT (effective to real address translation cache) when changing the PIDR register, which we do as part of guest entry and exit. Signed-off-by:
-
Paul Mackerras authored
If we allow LPCR[AIL] to be set for radix guests, then interrupts from the guest to the host can be delivered by the hardware with relocation on, and thus the code path starting at kvmppc_interrupt_hv can be executed in virtual mode (MMU on) for radix guests (previously it was only ever executed in real mode). Most of the code is indifferent to whether the MMU is on or off, but the calls to OPAL that use the real-mode OPAL entry code need to be switched to use the virtual-mode code instead. The affected calls are the calls to the OPAL XICS emulation functions in kvmppc_read_one_intr() and related functions. We test the MSR[IR] bit to detect whether we are in real or virtual mode, and call the opal_rm_* or opal_* function as appropriate. The other place that depends on the MMU being off is the optimization where the guest exit code jumps to the external interrupt vector or hypervisor doorbell interrupt vector, or returns to its caller (which is __kvmppc_vcore_entry). If the MMU is on and we are returning to the caller, then we don't need to use an rfid instruction since the MMU is already on; a simple blr suffices. If there is an external or hypervisor doorbell interrupt to handle, we branch to the relocation-on version of the interrupt vector. Signed-off-by:
-
Paul Mackerras authored
With radix, the guest can do TLB invalidations itself using the tlbie (global) and tlbiel (local) TLB invalidation instructions. Linux guests use local TLB invalidations for translations that have only ever been accessed on one vcpu. However, that doesn't mean that the translations have only been accessed on one physical cpu (pcpu) since vcpus can move around from one pcpu to another. Thus a tlbiel might leave behind stale TLB entries on a pcpu where the vcpu previously ran, and if that task then moves back to that previous pcpu, it could see those stale TLB entries and thus access memory incorrectly. The usual symptom of this is random segfaults in userspace programs in the guest. To cope with this, we detect when a vcpu is about to start executing on a thread in a core that is a different core from the last time it executed. If that is the case, then we mark the core as needing a TLB flush and then send an interrupt to any thread in the core that is currently running a vcpu from the same guest. This will get those vcpus out of the guest, and the first one to re-enter the guest will do the TLB flush. The reason for interrupting the vcpus executing on the old core is to cope with the following scenario: CPU 0 CPU 1 CPU 4 (core 0) (core 0) (core 1) VCPU 0 runs task X VCPU 1 runs core 0 TLB gets entries from task X VCPU 0 moves to CPU 4 VCPU 0 runs task X Unmap pages of task X tlbiel (still VCPU 1) task X moves to VCPU 1 task X runs task X sees stale TLB entries That is, as soon as the VCPU starts executing on the new core, it could unmap and tlbiel some page table entries, and then the task could migrate to one of the VCPUs running on the old core and potentially see stale TLB entries. Since the TLB is shared between all the threads in a core, we only use the bit of kvm->arch.need_tlb_flush corresponding to the first thread in the core. To ensure that we don't have a window where we can miss a flush, this moves the clearing of the bit from before the actual flush to after it. This way, two threads might both do the flush, but we prevent the situation where one thread can enter the guest before the flush is finished. Signed-off-by:
-
Paul Mackerras authored
If the guest is in radix mode, then it doesn't have a hashed page table (HPT), so all of the hypercalls that manipulate the HPT can't work and should return an error. This adds checks to make them return H_FUNCTION ("function not supported"). Signed-off-by: -
Paul Mackerras authored
This adds code to keep track of dirty pages when requested (that is, when memslot->dirty_bitmap is non-NULL) for radix guests. We use the dirty bits in the PTEs in the second-level (partition-scoped) page tables, together with a bitmap of pages that were dirty when their PTE was invalidated (e.g., when the page was paged out). This bitmap is stored in the first half of the memslot->dirty_bitmap area, and kvm_vm_ioctl_get_dirty_log_hv() now uses the second half for the bitmap that gets returned to userspace. Signed-off-by:
-
Paul Mackerras authored
This adapts our implementations of the MMU notifier callbacks (unmap_hva, unmap_hva_range, age_hva, test_age_hva, set_spte_hva) to call radix functions when the guest is using radix. These implementations are much simpler than for HPT guests because we have only one PTE to deal with, so we don't need to traverse rmap chains. Signed-off-by:
-
Paul Mackerras authored
This adds the code to construct the second-level ("partition-scoped" in architecturese) page tables for guests using the radix MMU. Apart from the PGD level, which is allocated when the guest is created, the rest of the tree is all constructed in response to hypervisor page faults. As well as hypervisor page faults for missing pages, we also get faults for reference/change (RC) bits needing to be set, as well as various other error conditions. For now, we only set the R or C bit in the guest page table if the same bit is set in the host PTE for the backing page. This code can take advantage of the guest being backed with either transparent or ordinary 2MB huge pages, and insert 2MB page entries into the guest page tables. There is no support for 1GB huge pages yet. Signed-off-by: -
Paul Mackerras authored
This adds code to branch around the parts that radix guests don't need - clearing and loading the SLB with the guest SLB contents, saving the guest SLB contents on exit, and restoring the host SLB contents. Since the host is now using radix, we need to save and restore the host value for the PID register. On hypervisor data/instruction storage interrupts, we don't do the guest HPT lookup on radix, but just save the guest physical address for the fault (from the ASDR register) in the vcpu struct. Signed-off-by:
-
Paul Mackerras authored
This adds a field in struct kvm_arch and an inline helper to indicate whether a guest is a radix guest or not, plus a new file to contain the radix MMU code, which currently contains just a translate function which knows how to traverse the guest page tables to translate an address. Signed-off-by:
-
Paul Mackerras authored
POWER9 adds a register called ASDR (Access Segment Descriptor Register), which is set by hypervisor data/instruction storage interrupts to contain the segment descriptor for the address being accessed, assuming the guest is using HPT translation. (For radix guests, it contains the guest real address of the access.) Thus, for HPT guests on POWER9, we can use this register rather than looking up the SLB with the slbfee. instruction. Signed-off-by:
-
Paul Mackerras authored
This adds the implementation of the KVM_PPC_CONFIGURE_V3_MMU ioctl for HPT guests on POWER9. With this, we can return 1 for the KVM_CAP_PPC_MMU_HASH_V3 capability. Signed-off-by:
-
Paul Mackerras authored
This adds two capabilities and two ioctls to allow userspace to find out about and configure the POWER9 MMU in a guest. The two capabilities tell userspace whether KVM can support a guest using the radix MMU, or using the hashed page table (HPT) MMU with a process table and segment tables. (Note that the MMUs in the POWER9 processor cores do not use the process and segment tables when in HPT mode, but the nest MMU does). The KVM_PPC_CONFIGURE_V3_MMU ioctl allows userspace to specify whether a guest will use the radix MMU or the HPT MMU, and to specify the size and location (in guest space) of the process table. The KVM_PPC_GET_RMMU_INFO ioctl gives userspace information about the radix MMU. It returns a list of supported radix tree geometries (base page size and number of bits indexed at each level of the radix tree) and the encoding used to specify the various page sizes for the TLB invalidate entry instruction. Initially, both capabilities return 0 and the ioctls return -EINVAL, until the necessary infrastructure for them to operate correctly is added. Signed-off-by:
-
Paul Mackerras authored
With host and guest both using radix translation, it is feasible for the host to take interrupts that come from the guest with relocation on, and that is in fact what the POWER9 hardware will do when LPCR[AIL] = 3. All such interrupts use HSRR0/1 not SRR0/1 except for system call with LEV=1 (hcall). Therefore this adds the KVM tests to the _HV variants of the relocation-on interrupt handlers, and adds the KVM test to the relocation-on system call entry point. We also instantiate the relocation-on versions of the hypervisor data storage and instruction interrupt handlers, since these can occur with relocation on in radix guests. Signed-off-by:
-
Paul Mackerras authored
When changing a partition table entry on POWER9, we do a particular form of the tlbie instruction which flushes all TLBs and caches of the partition table for a given logical partition ID (LPID). This instruction has a field in the instruction word, labelled R (radix), which should be 1 if the partition was previously a radix partition and 0 if it was a HPT partition. This implements that logic. Signed-off-by:
-
Paul Mackerras authored
This exports the pgtable_cache array and the pgtable_cache_add function so that HV KVM can use them for allocating radix page tables for guests. Signed-off-by:
-
Paul Mackerras authored
This adds definitions for bits in the DSISR register which are used by POWER9 for various translation-related exception conditions, and for some more bits in the partition table entry that will be needed by KVM. Signed-off-by:
-
Paul Mackerras authored
To use radix as a guest, we first need to tell the hypervisor via the ibm,client-architecture call first that we support POWER9 and architecture v3.00, and that we can do either radix or hash and that we would like to choose later using an hcall (the H_REGISTER_PROC_TBL hcall). Then we need to check whether the hypervisor agreed to us using radix. We need to do this very early on in the kernel boot process before any of the MMU initialization is done. If the hypervisor doesn't agree, we can't use radix and therefore clear the radix MMU feature bit. Later, when we have set up our process table, which points to the radix tree for each process, we need to install that using the H_REGISTER_PROC_TBL hcall. Signed-off-by:
-
Paul Mackerras authored
This fixes the byte index values for some of the option bits in the "ibm,architectur-vec-5" property. The "platform facilities options" bits are in byte 17 not byte 14, so the upper 8 bits of their definitions need to be 0x11 not 0x0E. The "sub processor support" option is in byte 21 not byte 15. Note none of these options are actually looked up in "ibm,architecture-vec-5" at this time, so there is no bug. When checking whether option bits are set, we should check that the offset of the byte being checked is less than the vector length that we got from the hypervisor. Signed-off-by:
-
Paul Mackerras authored
Currently, if the kernel is running on a POWER9 processor under a hypervisor, it will try to use the radix MMU even though it doesn't have the necessary code to use radix under a hypervisor (it doesn't negotiate use of radix, and it doesn't do the H_REGISTER_PROC_TBL hcall). The result is that the guest kernel will crash when it tries to turn on the MMU. This fixes it by looking for the /chosen/ibm,architecture-vec-5 property, and if it exists, clears the radix MMU feature bit, before we decide whether to initialize for radix or HPT. This property is created by the hypervisor as a result of the guest calling the ibm,client-architecture-support method to indicate its capabilities, so it will indicate whether the hypervisor agreed to us using radix. Systems without a hypervisor may have this property also (for example, skiboot creates it), so we check the HV bit in the MSR to see whether we are running as a guest or not. If we are in hypervisor mode, then we can do whatever we like including using the radix MMU. The reason for using this property is that in future, when we have support for using radix under a hypervisor, we will need to check this property to see whether the hypervisor agreed to us using radix. Fixes: 2bfd65e4 ("powerpc/mm/radix: Add radix callbacks for early init routines") Cc: stable@vger.kernel.org # v4.7+ Signed-off-by:
-
Nicholas Piggin authored
64-bit Book3S exception handlers must find the dynamic kernel base to add to the target address when branching beyond __end_interrupts, in order to support kernel running at non-0 physical address. Support this in KVM by branching with CTR, similarly to regular interrupt handlers. The guest CTR saved in HSTATE_SCRATCH1 and restored after the branch. Without this, the host kernel hangs and crashes randomly when it is running at a non-0 address and a KVM guest is started. Signed-off-by:
Nicholas Piggin <npiggin@gmail.com> Acked-by:
-
- 27 Jan, 2017 4 commits
-
-
Thomas Huth authored
The function kvmppc_handle_exit_pr() is quite huge and thus hard to read, and even contains a "spaghetti-code"-like goto between the different case labels of the big switch statement. This can be made much more readable by moving the code related to injecting program interrupts / instruction emulation into a separate function instead. Signed-off-by:
-
Paul Mackerras authored
The H_PROD hypercall is supposed to wake up an idle vcpu. We have an implementation, but because Linux doesn't use it except when doing cpu hotplug, it was never tested properly. AIX does use it, and reported it broken. It turns out we were waking the wrong vcpu (the one doing H_PROD, not the target of the prod) and we weren't handling the case where the target needs an IPI to wake it. Fix it by using the existing kvmppc_fast_vcpu_kick_hv() function, which is intended for this kind of thing, and by using the target vcpu not the current vcpu. We were also not looking at the prodded flag when checking whether a ceded vcpu should wake up, so this adds checks for the prodded flag alongside the checks for pending exceptions. Signed-off-by: -
Nicholas Piggin authored
A subsequent patch to make KVM handlers relocation-safe makes them unusable from within alt section "else" cases (due to the way fixed addresses are taken from within fixed section head code). Stop open-coding the KVM handlers, and add them both as normal. A more optimal fix may be to allow some level of alternate feature patching in the exception macros themselves, but for now this will do. The TRAMP_KVM handlers must be moved to the "virt" fixed section area (name is arbitrary) in order to be closer to .text and avoid the dreaded "relocation truncated to fit" error. Signed-off-by:
-
Nicholas Piggin authored
Change the calling convention to put the trap number together with CR in two halves of r12, which frees up HSTATE_SCRATCH2 in the HV handler. The 64-bit PR handler entry translates the calling convention back to match the previous call convention (i.e., shared with 32-bit), for simplicity. Signed-off-by:
-
- 26 Jan, 2017 3 commits
-
-
Li Zhong authored
This patch improves the code that takes lock twice to check the resend flag and do the actual resending, by checking the resend flag locklessly, and add a boolean parameter check_resend to icp_[rm_]deliver_irq(), so the resend flag can be checked in the lock when doing the delivery. We need make sure when we clear the ics's bit in the icp's resend_map, we don't miss the resend flag of the irqs that set the bit. It could be ordered through the barrier in test_and_clear_bit(), and a newly added wmb between setting irq's resend flag, and icp's resend_map. Signed-off-by:
Li Zhong <zhong@linux.vnet.ibm.com> Signed-off-by:
-
Li Zhong authored
This patch implements P(Presented)/Q(Queued) states for ICS irqs. When the interrupt is presented, set P. Present if P was not set. If P is already set, don't present again, set Q. When the interrupt is EOI'ed, move Q into P (and clear Q). If it is set, re-present. The asserted flag used by LSI is also incorporated into the P bit. When the irq state is saved, P/Q bits are also saved, they need some qemu modifications to be recognized and passed around to be restored. KVM_XICS_PENDING bit set and saved should also indicate KVM_XICS_PRESENTED bit set and saved. But it is possible some old code doesn't have/recognize the P bit, so when we restore, we set P for PENDING bit, too. The idea and much of the code come from Ben. Signed-off-by:
-
Li Zhong authored
It is possible that in the following order, one irq is resent twice: CPU 1 CPU 2 ics_check_resend() lock ics_lock see resend set unlock ics_lock /* change affinity of the irq */ kvmppc_xics_set_xive() write_xive() lock ics_lock see resend set unlock ics_lock icp_deliver_irq() /* resend */ icp_deliver_irq() /* resend again */ It doesn't have any user-visible effect at present, but needs to be avoided when the following patch implementing the P/Q stuff is applied. This patch clears the resend flag before releasing the ics lock, when we know we will do a re-delivery after checking the flag, or setting the flag. Signed-off-by:
-
